We had a similar problem on our webserver software with a memory leak. The process would hang, even though there was plenty of memory, process time, and disk time free.
We were stuck on the software, since the chunk that was causing it was a necessary evil from our software vendor, closed source, and untouchable.
We invented a similar watchdog sort of process to cycle the thing when it would do that. It ended up solving the problem about 99% of the time, so we ended up just calling it good.
Sometimes the effort required to chase it to that last 1%, if even possible, isn't worth it.
I think the only other thing that might be worth a shot is maybe mySQL on Linux on a test machine just to see if it has the same bug. I prefer Windows 2003 personally too, but I have a friend that does a lot of work with many different DBs, and he's always said that mySQL on Windows doesn't run quite the same as it does on Linux, unlike some of the other things that work the same on either OS.
The other thing we ran into was a perfectly good query that acted up just because of the version we were running at the time of Informix. When we tried to do a DATE -1 query to come up with yesterday's date (which worked on Access, MS SQL, mySQL, etc.), it would start something off that would make the DB just fall dead about 2-3 hours later. Reproduced it every time once we figured it out..... which took forever.
Still, sometimes something like simple process recyclers/killers solve the problem the easiest way.
